What matters in AI.

Subscribe

Claude Fable 5.2 leads an AI index at 65 but costs $26.54 a task

The cheapest model in the list costs $0.60 a task, so the top score costs about 44 times more.

Artificial Analysis updated its AI benchmarks (standard tests). Its Intelligence Index, which now uses nine tests, ranks Claude Fable 5.2 of Anthropic first at 65. Gemini 4 Carbon of Google follows at 60. The chart shows that the top model takes 41.8 minutes for a task. The fastest model takes 13.3 minutes. The cheapest model costs $0.60 for a task.

What is new

The company also updated two other benchmarks. Its API test uses time for each task as the primary measure and adds agentic work, where AI does tasks with many steps. Its Decision Index adds private tests and gives more points for correct confidence.

The numbers

Cost and time for each task, from the list:

  • Claude Fable 5.2: 65 points, $26.54, 41.8 minutes
  • GPT-6.1 Sol: 55 points, $1.27, 22.9 minutes
  • DeepSeek V4.1 Flash 1017: 33 points, $0.60, 13.3 minutes

What is not known

One score does not fit all tasks. Coding, cyber and decision tests measure different work. Some entries, such as Claude Fable 5.2, have a “No public API” label, so access is small. The page shows only 29 of 741 models.

Insight: The top score and the best price are different models

An earlier report said that Claude Fable 5.2 passed Gemini 4 Carbon but gave no scores. Meta released a model at $0.10 for each million input tokens. The new data shows a large difference in cost. The race for the top score seems to be a race that some buyers do not need to win, and power can move to cheap models that do the basic work. Monitor the distance to the leader.

Sources

  1. Artificial Analysis updates AI benchmarks with agentic tasks and calibrated-confidence scoringnews.lavx.hu
AI MATTER · NEWS · AI MATTER · NEWS ·16 NOV2026

Posted

Tags

More in Research

All Research news